In short, it's important to add this support as it can be easy for human misprocessing due to visual similarities between advisory which isn't legally binding and max speed which is legally binding resulting in advisory signs being being added to OSM as max speed limit which people may then act upon in OSMAnd+, etc.
Edit: I don't have background in machine learning, but I believe there is a valid risk in that the longer this support is put off, the greater the chance of advisory being trained in validation as max speed by humans (due to the current UX not providing any guidance such as Do's/Don'ts for any particular sign type such as don't validate x as y) which will make initial training harder to do whenever advisory support validation is added
Add image recognition for advisory speed limits.
In short, it's important to add this support as it can be easy for human misprocessing due to visual similarities between advisory which isn't legally binding and max speed which is legally binding resulting in advisory signs being being added to OSM as max speed limit which people may then act upon in OSMAnd+, etc.
Edit: I don't have background in machine learning, but I believe there is a valid risk in that the longer this support is put off, the greater the chance of advisory being trained in validation as max speed by humans (due to the current UX not providing any guidance such as Do's/Don'ts for any particular sign type such as don't validate x as y) which will make initial training harder to do whenever advisory support validation is added